Music Recommendation Engine
Second solo hackathon entry for Google's Gemini Long Context Competition. Open source project.
- Using Gemini API to demonstrate 1m token window and context caching features of Gemini Flash AI model for a unique use case.
- Created dataset from BigQuery to extract around 4 years of listening history for a specific user called SavageCore.
- Created cache of listening history and asked a number of questions to better understand SavageCore's music taste and to provide: tailored recommendations, auto playlist creation and to generate a musical personality.
- Introduced multiple users through separate caches, generated and saved unique musical profiles to variables.
- Used musical profiles to generate compatibility scores and user matching based on music taste.
- Added ability to upload a song and recommend a user that would be most likely to enjoy it based on their music profile.